Temperature prediction of heat sources using machine learning techniques

Abstract This paper explores the use of machine learning algorithms, such as XGBoost, random forest regression, support vector machine regression, and artificial neural network (ANN), which are employed for predicting temperatures of rectangular silicon heaters with dummy elements. A combination of these machine learning algorithms can predict better results over individual algorithm. Silicon heaters are equipped on an FR4 substrate board for cooling under forced convection in a horizontal channel. COMSOL Multiphysics 5.4 software is used for all the three‐dimensional numerical simulations. Heat transfer at the solid and fluid interface is studied using a module based on conjugate heat transfer and nonisothermal fluid flow. Dummy elements are coupled with heated sources to evaluate heat transfer and analyze the flow of fluid. The study is performed with 2.5 m/s velocity and a uniform heat flux of 5000 W/m 2 . The study is aimed at predicting and comparing results of support vector regression (SVR), ensemble learning with ANN to explore optimal configuration. Results indicate an agreement of less than 10% between the simulated and predicted temperatures. It is also found that SVR has given the best results compared with XG Boot and ANN when analyzed individually. The programming for these algorithms is performed using the Python programming language.